# Spam label checks for AI voice and sales agents

**A spam label is a silent failure for an AI voice agent: the call connects normally in your logs, but the recipient sees a warning like "Spam Risk" and doesn't answer.** Spam I Am lets you check what real phones on AT&T, Verizon and T-Mobile display when your agent's number calls them. Run a check when you provision a number, before each campaign, and on a schedule while the number is active.

*Covers AT&T, Verizon and T-Mobile. [See coverage](/coverage)*

*Start checks and read results from your own code with the [API](/docs/api). The first check on a number needs no key. Your calling platform places the test calls. Agents that use MCP can connect to our [MCP server](/docs/mcp).*

## Why AI voice calls get flagged

The same things that flag any caller hit AI agents especially hard:

- **Dead air at the start.** Hiya grades numbers partly on whether recipients stay on the line, and notes that people hang up quickly when there's silence after they say "hello." Response latency in a voice agent produces exactly that.
- **Bursty, short calls.** The FCC's examples of reasonable call analytics include large bursts of calls in a short window, low average call duration, and low completion rates. That's what an agent produces when it's pointed at a list and run at full speed.
- **Unexpected calls.** Hiya points to stale lead lists, minimal consent, and repeat calls after someone shows no interest as common causes of poor reputation.
- **Fresh numbers.** Numbers with little calling history can draw extra scrutiny. [Checklist for new numbers](/guides/check-new-number-before-dialer)

### Consent matters more for AI calls
In February 2024, the FCC ruled (FCC 24-17) that AI-generated voices count as "artificial or prerecorded voice" under the TCPA. So AI-voice calls generally need the called party's prior express consent unless an exemption applies. Calls people didn't agree to get reported, and reports drive labels. *This isn't legal advice. Talk to counsel about your use case.*

## What your logs can and can't tell you

| Signal | What it means | Visible in your logs? |
|---|---|---|
| Call blocked by the carrier | Analytics stopped the call. The phone never rang. | Yes, as SIP 603+ "Network Blocked." [What it means](/guides/sip-603-plus-608-network-blocked) |
| Call labeled ("Spam Risk," "Scam Likely," etc.) | The phone rang with a warning. | **No.** It looks like a normal unanswered call. |
| Call silenced by an app on the phone | The recipient's device screened it. | Usually no |
| Answer rate dropping | Something is wrong. | Yes, but it doesn't tell you what |

A label is the gap: the only way to see it is to look at the recipient's screen. That's what a check does.

## How a check works in your pipeline

1. **Start a check** for the number your agent calls from, with one API call (`POST /v1/checks`) or on the website.
2. **Place a call from that number to each carrier's test line** (one fixed line per carrier). The call must come from the number being checked. We never spoof numbers, so your platform has to originate it.
3. **Get the result:** the label (or none), the caller name, and a screenshot from a real phone on each carrier you checked.

- Typical time to a result: about 5 seconds
- Result delivery: from the API (`GET /v1/checks/{id}`, asked every few seconds), or on the check page, as each carrier's result arrives.
- The test call: our phone lets it ring for a few seconds, then sends it to voicemail, so your platform can hang up. It's an ordinary short call, billed by your provider like any other.
- Dial within 2 minutes for one carrier, plus 1 minute for each additional carrier. If no call arrives, the result is Incomplete, at no charge.
- Limit: one check per number per hour.

## When to run checks

- **When a number is provisioned.** Don't release it to campaigns until it comes back clean, or a human decides what to do.
- **Before each campaign or list.** A quick pre-flight on every number you'll use.
- **On a schedule while numbers are active.** Labels can appear after a burst of volume or complaints.
- **After a volume spike or a drop in answer rates.**
- **Every few days while a label is under review,** to see when it clears.

## When a check comes back flagged

1. **Pause calls from that number.** More calls from a flagged number usually mean more unanswered calls and more reports.
2. **Alert a human.** A person should decide what changes and file the review.
3. **Fix the cause:** latency, volume, list quality, consent. [Why numbers get flagged](/guides/why-number-shows-spam-likely)
4. **Request a review** from that carrier's analytics provider: [AT&T](/guides/remove-spam-label-att) · [Verizon](/guides/remove-spam-label-verizon) · [T-Mobile](/guides/remove-scam-likely-tmobile)
5. **Re-check every few days** until it clears.

**Incomplete, even though you dialed from the right number with caller ID on?** The call may have been blocked before it rang. Treat it as more urgent than a label, and check your logs for SIP 603+.

**Don't auto-rotate.** Swapping in a fresh number whenever one gets flagged hides the problem from you, and the behavior looks like label evasion to the analytics engines. We won't help build that, and it doesn't fix anything.

## Checks your agent can run itself

An agent that notices low answer rates or call rejections can check its own number:

- **The API is live.** The first check on a number needs no key, so an agent can try it mid-task. [API docs](/docs/api)
- **Approval by a human.** When the agent needs to spend, it sends its person a link. They sign in, add funds if needed, set a spending limit and approve. The agent then gets its own key, capped at that limit.
- **Webhooks.** Results can be sent to your server when a check finishes, instead of your agent asking for them.
- **An MCP server** for agents that use Model Context Protocol, at `https://ismynumberspam.com/mcp`. [MCP docs](/docs/mcp)

Plain-language notes for agents: [llms.txt](https://ismynumberspam.com/llms.txt)

## Pricing

- **First check on any number: free**, no key or signup
- **$0.25 for one carrier, or $0.40 for two or three,** from a prepaid balance. Add funds from $10; they never expire.
- **Coming soon: $45/month** for unlimited checks, for high-volume pipelines (each number can still be checked once an hour)

No subscription required and no sales call. [Pricing](/pricing)

## Limits

- **Three carriers.** We test on AT&T, Verizon and T-Mobile. A clean result on one says nothing about the others, so check all three.
- **A snapshot.** One call, one device, one moment. Re-check over time.
- **Checking only.** We don't monitor numbers for you unless you schedule checks, and we don't remove labels.

## FAQ

### Why are my AI agent's calls going unanswered when my logs show no errors?
They may be labeled. A call that rings with "Spam Risk" or "Scam Likely" on screen produces no error code, so your logs show a normal unanswered call. Only blocked calls return SIP 603+. Check what the recipient's screen actually shows.

### Do AI voice calls need consent?
Under a February 2024 FCC ruling, AI-generated voices count as "artificial or prerecorded voice" under the TCPA, so such calls generally require the called party's prior express consent unless an exemption applies. Consult a lawyer about your specific use case.

### How often should I check my agent's numbers?
At provisioning, before each campaign, on a regular schedule while numbers are active, and every few days while a label is under review.

### Can I check a number my agent doesn't call from?
No. The check call has to come from the number being checked, because we never spoof numbers. In practice, that means you can only check numbers you control.

### What should my pipeline do when a number is flagged?
Pause calls from it, alert a human, fix the cause, and request a review. Don't automatically rotate to a new number.

## Sources

- [Hiya Connect: Caller reputation and spam label help](https://connect-support.hiya.com/hc/en-us/articles/54311390207891-Caller-reputation-and-spam-label-help)
- [ACA International: Summary of FCC 2019 declaratory ruling](https://policymakers.acainternational.org/wp-content/uploads/2020/09/concerns-regarding-fcc-call-blocking-and-labeling-efforts-declaratory-ruling-and-fnprm-may-2019.pdf)
- [TransNexus: FCC ruling on AI-generated voice calls](https://transnexus.com/blog/2024/ai-voice-calls-and-tcpa/)
- [Mayer Brown: FCC declares authority to regulate AI-generated calls under the TCPA](https://www.mayerbrown.com/en/insights/publications/2024/02/fcc-declares-authority-and-intent-to-regulate-ai-generated-calls-under-the-tcpa)
- [CompliancePoint: What SIP 603+ means for telemarketing](https://www.compliancepoint.com/marketing-compliance/sip-603-what-the-fccs-new-call-blocking-rule-means-for-telemarketing/)
- [Telecom Law Partners: Eighth Robocall R&O published](https://www.tlp.law/2025/03/26/fcc-publishes-eighth-robocall-ro-in-federal-register/)